478 lines
22 KiB
Python
Executable File
478 lines
22 KiB
Python
Executable File
#!/usr/bin/env python3
|
||
|
||
|
||
import argparse, re, random, string
|
||
from scipy.constants import hbar, e as eV, pi, c
|
||
|
||
def make_action_sharedlist(opname, listname):
|
||
class opAction(argparse.Action):
|
||
def __call__(self, parser, args, values, option_string=None):
|
||
if (not hasattr(args, listname)) or getattr(args, listname) is None:
|
||
setattr(args, listname, list())
|
||
getattr(args,listname).append((opname, values))
|
||
return opAction
|
||
|
||
parser = argparse.ArgumentParser()
|
||
#TODO? použít type=argparse.FileType('r') ?
|
||
parser.add_argument('--TMatrix', action='store', required=True, help='Path to TMatrix file')
|
||
parser.add_argument('--griddir', action='store', required=True, help='Path to the directory with precalculated translation operators')
|
||
#sizepar = parser.add_mutually_exclusive_group(required=True)
|
||
parser.add_argument('--hexside', action='store', type=float, required=True, help='Lattice hexagon size length')
|
||
parser.add_argument('--output', action='store', help='Path to output PDF')
|
||
parser.add_argument('--nSV', action='store', metavar='N', type=int, default=1, help='Store and draw N minimun singular values')
|
||
parser.add_argument('--background_permittivity', action='store', type=float, default=1., help='Background medium relative permittivity (default 1)')
|
||
parser.add_argument('--sparse', action='store', type=int, help='Skip frequencies for preview')
|
||
parser.add_argument('--eVmax', action='store', type=float, help='Skip frequencies above this value')
|
||
parser.add_argument('--eVmin', action='store', type=float, help='Skip frequencies below this value')
|
||
parser.add_argument('--kdensity', action='store', type=int, default=66, help='Number of k-points per x-axis segment')
|
||
#TODO some more sophisticated x axis definitions
|
||
parser.add_argument('--gaussian', action='store', type=float, metavar='σ', help='Use a gaussian envelope for weighting the interaction matrix contributions (depending on the distance), measured in unit cell lengths (?) FIxME).')
|
||
popgrp=parser.add_argument_group(title='Operations')
|
||
popgrp.add_argument('--tr', dest='ops', action=make_action_sharedlist('tr', 'ops'), default=list()) # the default value for dest can be set once
|
||
popgrp.add_argument('--tr0', dest='ops', action=make_action_sharedlist('tr0', 'ops'))
|
||
popgrp.add_argument('--tr1', dest='ops', action=make_action_sharedlist('tr1', 'ops'))
|
||
popgrp.add_argument('--sym', dest='ops', action=make_action_sharedlist('sym', 'ops'))
|
||
popgrp.add_argument('--sym0', dest='ops', action=make_action_sharedlist('sym0', 'ops'))
|
||
popgrp.add_argument('--sym1', dest='ops', action=make_action_sharedlist('sym1', 'ops'))
|
||
#popgrp.add_argument('--mult', dest='ops', nargs=3, metavar=('INCSPEC', 'SCATSPEC', 'MULTIPLIER'), action=make_action_sharedlist('mult', 'ops'))
|
||
#popgrp.add_argument('--mult0', dest='ops', nargs=3, metavar=('INCSPEC', 'SCATSPEC', 'MULTIPLIER'), action=make_action_sharedlist('mult0', 'ops'))
|
||
#popgrp.add_argument('--mult1', dest='ops', nargs=3, metavar=('INCSPEC', 'SCATSPEC', 'MULTIPLIER'), action=make_action_sharedlist('mult1', 'ops'))
|
||
popgrp.add_argument('--multl', dest='ops', nargs=3, metavar=('INCL[,INCL,...]', 'SCATL[,SCATL,...]', 'MULTIPLIER'), action=make_action_sharedlist('multl', 'ops'))
|
||
popgrp.add_argument('--multl0', dest='ops', nargs=3, metavar=('INCL[,INCL,...]', 'SCATL[,SCATL,...]', 'MULTIPLIER'), action=make_action_sharedlist('multl0', 'ops'))
|
||
popgrp.add_argument('--multl1', dest='ops', nargs=3, metavar=('INCL[,INCL,...]', 'SCATL[,SCATL,...]', 'MULTIPLIER'), action=make_action_sharedlist('multl1', 'ops'))
|
||
parser.add_argument('--frequency_multiplier', action='store', type=float, default=1., help='Multiplies the frequencies in the TMatrix file by a given factor.')
|
||
# TODO enable more flexible per-sublattice specification
|
||
pargs=parser.parse_args()
|
||
print(pargs)
|
||
|
||
|
||
translations_dir = pargs.griddir
|
||
TMatrix_file = pargs.TMatrix
|
||
pdfout = pargs.output if pargs.output else (''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(10)) + '.pdf')
|
||
print(pdfout)
|
||
|
||
hexside = pargs.hexside #375e-9
|
||
epsilon_b = pargs.background_permittivity #2.3104
|
||
gaussianSigma = pargs.gaussian if pargs.gaussian else None # hexside * 222 / 7
|
||
interpfreqfactor = pargs.frequency_multiplier
|
||
kdensity = pargs.kdensity
|
||
minfreq = pargs.eVmin*eV/hbar if pargs.eVmin else None
|
||
maxfreq = pargs.eVmax*eV/hbar if pargs.eVmax else None
|
||
skipfreq = pargs.sparse if pargs.sparse else None
|
||
svn = pargs.nSV
|
||
|
||
# TODO multiplier operation definitions and parsing
|
||
#factor13inc = 10
|
||
#factor13scat=10
|
||
|
||
ops = list()
|
||
opre = re.compile('(tr|sym|copy|multl|mult)(\d*)')
|
||
for oparg in pargs.ops:
|
||
opm = opre.match(oparg[0])
|
||
if opm:
|
||
ops.append(((opm.group(2),) if opm.group(2) else (0,1), opm.group(1), oparg[1]))
|
||
else:
|
||
raise # should not happen
|
||
print(ops)
|
||
|
||
#ops = (
|
||
# # co, typ operace (symetrizace / transformace / kopie), specifikace (operace nebo zdroj),
|
||
# # co: 0, 1, (0,1), (0,), (1,), #NI: 'all'
|
||
# # typ operace: sym, tr, copy
|
||
# # specifikace:
|
||
# # sym, tr: 'σ_z', 'σ_y', 'C2'; sym: 'C3',
|
||
# # copy: 0, 1 (zdroj)
|
||
# ((0,1), 'sym', 'σ_z'),
|
||
# #((0,1), 'sym', 'σ_x'),
|
||
# #((0,1), 'sym', 'σ_y'),
|
||
# ((0,1), 'sym', 'C3'),
|
||
# ((1), 'tr', 'C2'),
|
||
#
|
||
#)
|
||
|
||
# -----------------finished basic CLI parsing (except for op arguments) ------------------
|
||
import time
|
||
begtime=time.time()
|
||
|
||
from matplotlib.path import Path
|
||
import matplotlib.patches as patches
|
||
import matplotlib.pyplot as plt
|
||
import qpms
|
||
import numpy as np
|
||
import os, sys, warnings, math
|
||
from matplotlib import pyplot as plt
|
||
from matplotlib.backends.backend_pdf import PdfPages
|
||
from scipy import interpolate
|
||
nx = None
|
||
s3 = math.sqrt(3)
|
||
|
||
|
||
|
||
pdf = PdfPages(pdfout)
|
||
|
||
# In[3]:
|
||
|
||
# specifikace T-matice zde
|
||
cdn = c/ math.sqrt(epsilon_b)
|
||
TMatrices_orig, freqs_orig, freqs_weirdunits_orig, lMax = qpms.loadScuffTMatrices(TMatrix_file)
|
||
my, ny = qpms.get_mn_y(lMax)
|
||
nelem = len(my)
|
||
|
||
ž = np.arange(2*nelem)
|
||
tž = ž // nelem
|
||
mž = my[ž%nelem]
|
||
nž = ny[ž%nelem]
|
||
TEž = ž[(mž+nž+tž) % 2 == 0]
|
||
TMž = ž[(mž+nž+tž) % 2 == 1]
|
||
|
||
č = np.arange(2*2*nelem)
|
||
žč = č % (2* nelem)
|
||
tč = tž[žč]
|
||
mč = mž[žč]
|
||
nč = nž[žč]
|
||
TEč = č[(mč+nč+tč) % 2 == 0]
|
||
TMč = č[(mč+nč+tč) % 2 == 1]
|
||
|
||
TMatrices = np.array(np.broadcast_to(TMatrices_orig[:,nx,:,:,:,:],(len(freqs_orig),2,2,nelem,2,nelem)) )
|
||
|
||
#TMatrices[:,:,:,:,:,ny==3] *= factor13inc
|
||
#TMatrices[:,:,:,ny==3,:,:] *= factor13scat
|
||
xfl = qpms.xflip_tyty(lMax)
|
||
yfl = qpms.yflip_tyty(lMax)
|
||
zfl = qpms.zflip_tyty(lMax)
|
||
c2rot = qpms.apply_matrix_left(qpms.yflip_yy(3),qpms.xflip_yy(3),-1)
|
||
|
||
reCN = re.compile('(\d*)C(\d+)')
|
||
#TODO C nekonečno
|
||
|
||
for op in ops:
|
||
if op[0] == 'all':
|
||
targets = (0,1)
|
||
elif isinstance(op[0],int):
|
||
targets = (op[0],)
|
||
else:
|
||
targets = op[0]
|
||
|
||
if op[1] == 'sym':
|
||
mCN = reCN.match(op[2]) # Fuck van Rossum for not having assignments inside expressions
|
||
if op[2] == 'σ_z':
|
||
for t in targets:
|
||
TMatrices[:,t] = (TMatrices[:,t] + qpms.apply_ndmatrix_left(zfl,qpms.apply_ndmatrix_left(zfl, TMatrices[:,t], (-4,-3)),(-2,-1)))/2
|
||
elif op[2] == 'σ_y':
|
||
for t in targets:
|
||
TMatrices[:,t] = (TMatrices[:,t] + qpms.apply_ndmatrix_left(yfl,qpms.apply_ndmatrix_left(yfl, TMatrices[:,t], (-4,-3)),(-2,-1)))/2
|
||
elif op[2] == 'σ_x':
|
||
for t in targets:
|
||
TMatrices[:,t] = (TMatrices[:,t] + qpms.apply_ndmatrix_left(xfl,qpms.apply_ndmatrix_left(xfl, TMatrices[:,t], (-4,-3)),(-2,-1)))/2
|
||
elif op[2] == 'C2': # special case of the latter
|
||
for t in targets:
|
||
TMatrices[:,t] = (TMatrices[:,t] + qpms.apply_matrix_left(c2rot,qpms.apply_matrix_left(c2rot, TMatrices[:,t], -3),-1))/2
|
||
elif mCN:
|
||
rotN = int(mCN.group(2))
|
||
TMatrix_contribs = np.empty((rotN,TMatrices.shape[0],2,nelem,2,nelem), dtype=np.complex_)
|
||
for t in targets:
|
||
for i in range(rotN):
|
||
rotangle = 2*np.pi*i / rotN
|
||
rot = qpms.WignerD_yy_fromvector(lMax,np.array([0,0,rotangle]))
|
||
rotinv = qpms.WignerD_yy_fromvector(lMax,np.array([0,0,-rotangle]))
|
||
TMatrix_contribs[i] = qpms.apply_matrix_left(rot,qpms.apply_matrix_left(rotinv, TMatrices[:,t], -3),-1)
|
||
TMatrices[:,t] = np.sum(TMatrix_contribs, axis=0) / rotN
|
||
else:
|
||
raise
|
||
elif op[1] == 'tr':
|
||
mCN = reCN.match(op[2]) # Fuck van Rossum for not having assignments inside expressions
|
||
if op[2] == 'σ_z':
|
||
for t in targets:
|
||
TMatrices[:,t] = qpms.apply_ndmatrix_left(zfl,qpms.apply_ndmatrix_left(zfl, TMatrices[:,t], (-4,-3)),(-2,-1))
|
||
elif op[2] == 'σ_y':
|
||
for t in targets:
|
||
TMatrices[:,t] = qpms.apply_ndmatrix_left(yfl,qpms.apply_ndmatrix_left(yfl, TMatrices[:,t], (-4,-3)),(-2,-1))
|
||
elif op[2] == 'σ_x':
|
||
for t in targets:
|
||
TMatrices[:,t] = qpms.apply_ndmatrix_left(xfl,qpms.apply_ndmatrix_left(xfl, TMatrices[:,t], (-4,-3)),(-2,-1))
|
||
elif op[2] == 'C2':
|
||
for t in targets:
|
||
TMatrices[:,t] = qpms.apply_matrix_left(c2rot,qpms.apply_matrix_left(c2rot, TMatrices[:,t], -3),-1)
|
||
elif mCN:
|
||
rotN = int(mCN.group(2))
|
||
power = int(mCN.group(1)) if mCN.group(1) else 1
|
||
TMatrix_contribs = np.empty((rotN,TMatrices.shape[0],2,nelem,2,nelem), dtype=np.complex_)
|
||
for t in targets:
|
||
rotangle = 2*np.pi*power/rotN
|
||
rot = qpms.WignerD_yy_fromvector(lMax, np.array([0,0,rotangle]))
|
||
rotinv = qpms.WignerD_yy_fromvector(lMax, np.array([0,0,-rotangle]))
|
||
TMatrices[:,t] = qpms.apply_matrix_left(rot, qpms.apply_matrix_left(rotinv, TMatrices[:,t], -3),-1)
|
||
else:
|
||
raise
|
||
elif op[1] == 'copy':
|
||
raise # not implemented
|
||
elif op[1] == 'mult':
|
||
raise # not implemented
|
||
elif op[1] == 'multl':
|
||
incy = np.full((nelem,), False, dtype=bool)
|
||
for incl in op[2][0].split(','):
|
||
l = int(incl)
|
||
incy += (l == ny)
|
||
scaty = np.full((nelem,), False, dtype=bool)
|
||
for scatl in op[2][1].split(','):
|
||
l = int(scatl)
|
||
scaty += (l == ny)
|
||
for t in targets:
|
||
TMatrices[:,t,:,scaty,:,incy] *= float(op[2][2])
|
||
else:
|
||
raise #unknown operation; should not happen
|
||
|
||
TMatrices_interp = interpolate.interp1d(freqs_orig*interpfreqfactor, TMatrices, axis=0, kind='linear',fill_value="extrapolate")
|
||
|
||
|
||
|
||
# In[4]:
|
||
|
||
om = np.linspace(np.min(freqs_orig), np.max(freqs_orig),100)
|
||
TMatrix0ip = np.reshape(TMatrices_interp(om)[:,0], (len(om), 2*nelem*2*nelem))
|
||
f, axa = plt.subplots(2, 2, figsize=(15,15))
|
||
|
||
#print(TMatrices.shape)
|
||
#plt.plot(om, TMatrices[:,0,0,0,0].imag,'r',om, TMatrices[:,0,0,0,0].real,'r--',om, TMatrices[:,0,2,0,2].imag,'b',om, TMatrices[:,0,2,0,2].real,'b--'))
|
||
|
||
ax = axa[0,0]
|
||
ax2 = ax.twiny()
|
||
ax2.set_xlim([ax.get_xlim()[0]/eV*hbar,ax.get_xlim()[1]/eV*hbar])
|
||
ax.plot(
|
||
om, TMatrix0ip[:,:].imag,'-',om, TMatrix0ip[:,:].real,'--',
|
||
)
|
||
ax = axa[0,1]
|
||
ax2 = ax.twiny()
|
||
ax2.set_xlim([ax.get_xlim()[0]/eV*hbar,ax.get_xlim()[1]/eV*hbar])
|
||
ax.plot(
|
||
om, abs(TMatrix0ip[:,:]),'-'
|
||
)
|
||
ax.set_yscale('log')
|
||
|
||
ax = axa[1,1]
|
||
ax2 = ax.twiny()
|
||
ax2.set_xlim([ax.get_xlim()[0]/eV*hbar,ax.get_xlim()[1]/eV*hbar])
|
||
ax.plot(
|
||
om, np.unwrap(np.angle(TMatrix0ip[:,:]),axis=0),'-'
|
||
)
|
||
|
||
ax = axa[1,0]
|
||
ax.text(0.5,0.5,str(pargs).replace(',',',\n'),horizontalalignment='center',verticalalignment='center',transform=ax.transAxes)
|
||
pdf.savefig(f)
|
||
|
||
|
||
# In[ ]:
|
||
|
||
#kdensity = 66 #defined from cl arguments
|
||
bz_0 = np.array((0,0,0.,))
|
||
bz_K1 = np.array((1.,0,0))*4*np.pi/3/hexside/s3
|
||
bz_K2 = np.array((1./2.,s3/2,0))*4*np.pi/3/hexside/s3
|
||
bz_M = np.array((3./4, s3/4,0))*4*np.pi/3/hexside/s3
|
||
k0Mlist = bz_0 + (bz_M-bz_0) * np.linspace(0,1,kdensity)[:,nx]
|
||
kMK1list = bz_M + (bz_K1-bz_M) * np.linspace(0,1,kdensity)[:,nx]
|
||
kK10list = bz_K1 + (bz_0-bz_K1) * np.linspace(0,1,kdensity)[:,nx]
|
||
k0K2list = bz_0 + (bz_K2-bz_0) * np.linspace(0,1,kdensity)[:,nx]
|
||
kK2Mlist = bz_K2 + (bz_M-bz_K2) * np.linspace(0,1,kdensity)[:,nx]
|
||
B1 = 2* bz_K1 - bz_K2
|
||
B2 = 2* bz_K2 - bz_K1
|
||
klist = np.concatenate((k0Mlist,kMK1list,kK10list,k0K2list,kK2Mlist), axis=0)
|
||
kxmaplist = np.concatenate((np.array([0]),np.cumsum(np.linalg.norm(np.diff(klist, axis=0), axis=-1))))
|
||
|
||
|
||
# In[ ]:
|
||
|
||
n2id = np.identity(2*nelem)
|
||
n2id.shape = (2,nelem,2,nelem)
|
||
extlistlist = list()
|
||
leftmatrixlistlist = list()
|
||
minsvTElistlist=list()
|
||
minsvTMlistlist=list()
|
||
nan = float('nan')
|
||
omegalist = list()
|
||
filecount = 0
|
||
for trfile in os.scandir(translations_dir):
|
||
filecount += 1
|
||
if (skipfreq and filecount % skipfreq):
|
||
continue
|
||
try:
|
||
npz = np.load(trfile.path, mmap_mode='r')
|
||
k_0 = npz['precalc_params'][()]['k_hexside'] / hexside
|
||
omega = k_0 * c / math.sqrt(epsilon_b)
|
||
if((minfreq and omega < minfreq) or (maxfreq and omega > maxfreq)):
|
||
continue
|
||
except:
|
||
print ("Unexpected error, trying to continue with another file:", sys.exc_info()[0])
|
||
continue
|
||
try:
|
||
tdic = qpms.hexlattice_precalc_AB_loadunwrap(trfile.path, return_points=True)
|
||
except:
|
||
print ("Unexpected error, trying to continue with another file:", sys.exc_info()[0])
|
||
continue
|
||
k_0 = tdic['k_hexside'] / hexside
|
||
omega = k_0 * c / math.sqrt(epsilon_b)
|
||
omegalist.append(omega)
|
||
print(filecount, omega/eV*hbar)
|
||
sys.stdout.flush()
|
||
a_self = tdic['a_self'][:,:nelem,:nelem]
|
||
b_self = tdic['b_self'][:,:nelem,:nelem]
|
||
a_u2d = tdic['a_u2d'][:,:nelem,:nelem]
|
||
b_u2d = tdic['b_u2d'][:,:nelem,:nelem]
|
||
a_d2u = tdic['a_d2u'][:,:nelem,:nelem]
|
||
b_d2u = tdic['b_d2u'][:,:nelem,:nelem]
|
||
unitcell_translations = tdic['self_tr']*hexside*s3
|
||
u2d_translations = tdic['u2d_tr']*hexside*s3
|
||
d2u_translations = tdic['d2u_tr']*hexside*s3
|
||
if gaussianSigma:
|
||
unitcell_envelope = np.exp(-np.sum(tdic['self_tr']**2,axis=-1)/(2*gaussianSigma**2))
|
||
u2d_envelope = np.exp(-np.sum(tdic['u2d_tr']**2,axis=-1)/(2*gaussianSigma**2))
|
||
d2u_envelope = np.exp(-np.sum(tdic['d2u_tr']**2,axis=-1)/(2*gaussianSigma**2))
|
||
|
||
|
||
TMatrices_om = TMatrices_interp(omega)
|
||
|
||
minsvTElist = np.full((klist.shape[0], svn),np.nan)
|
||
minsvTMlist = np.full((klist.shape[0], svn),np.nan)
|
||
leftmatrixlist = np.full((klist.shape[0],2,2,nelem,2,2,nelem),np.nan,dtype=complex)
|
||
isNaNlist = np.zeros((klist.shape[0]), dtype=bool)
|
||
|
||
# sem nějaká rozumná smyčka
|
||
for ki in range(klist.shape[0]):
|
||
k = klist[ki]
|
||
if (k_0*k_0 - k[0]*k[0] - k[1]*k[1] < 0):
|
||
isNaNlist[ki] = True
|
||
continue
|
||
|
||
phases_self = np.exp(1j*np.tensordot(k,unitcell_translations,axes=(0,-1)))
|
||
phases_u2d = np.exp(1j*np.tensordot(k,u2d_translations,axes=(0,-1)))
|
||
phases_d2u = np.exp(1j*np.tensordot(k,d2u_translations,axes=(0,-1)))
|
||
if gaussianSigma:
|
||
phases_self *= unitcell_envelope
|
||
phases_u2d *= u2d_envelope
|
||
phases_d2u *= d2u_envelope
|
||
leftmatrix = np.zeros((2,2,nelem, 2,2,nelem), dtype=complex)
|
||
|
||
leftmatrix[0,0,:,0,0,:] = np.tensordot(a_self,phases_self, axes=(0,-1)) # u2u, E2E
|
||
leftmatrix[1,0,:,1,0,:] = leftmatrix[0,0,:,0,0,:] # d2d, E2E
|
||
leftmatrix[0,1,:,0,1,:] = leftmatrix[0,0,:,0,0,:] # u2u, M2M
|
||
leftmatrix[1,1,:,1,1,:] = leftmatrix[0,0,:,0,0,:] # d2d, M2M
|
||
leftmatrix[0,0,:,0,1,:] = np.tensordot(b_self,phases_self, axes=(0,-1)) # u2u, M2E
|
||
leftmatrix[0,1,:,0,0,:] = leftmatrix[0,0,:,0,1,:] # u2u, E2M
|
||
leftmatrix[1,1,:,1,0,:] = leftmatrix[0,0,:,0,1,:] # d2d, E2M
|
||
leftmatrix[1,0,:,1,1,:] = leftmatrix[0,0,:,0,1,:] # d2d, M2E
|
||
leftmatrix[0,0,:,1,0,:] = np.tensordot(a_d2u, phases_d2u,axes=(0,-1)) #d2u,E2E
|
||
leftmatrix[0,1,:,1,1,:] = leftmatrix[0,0,:,1,0,:] #d2u, M2M
|
||
leftmatrix[0,0,:,1,1,:] = np.tensordot(b_d2u, phases_d2u,axes=(0,-1)) #d2u,M2E
|
||
leftmatrix[0,1,:,1,0,:] = leftmatrix[0,0,:,1,1,:] #d2u, E2M
|
||
leftmatrix[1,0,:,0,1,:] = np.tensordot(b_u2d, phases_u2d,axes=(0,-1)) #u2d,M2E
|
||
leftmatrix[1,1,:,0,0,:] = leftmatrix[1,0,:,0,1,:] #u2d, E2M
|
||
#leftmatrix is now the translation matrix T
|
||
for j in range(2):
|
||
leftmatrix[j] = -np.tensordot(TMatrices_om[j], leftmatrix[j], axes=([-2,-1],[0,1]))
|
||
# at this point, jth row of leftmatrix is that of -MT
|
||
leftmatrix[j,:,:,j,:,:] += n2id
|
||
#now we are done, 1-M
|
||
|
||
leftmatrixlist[ki] = leftmatrix
|
||
|
||
|
||
nnlist = np.logical_not(isNaNlist)
|
||
leftmatrixlist_s = np.reshape(leftmatrixlist,(klist.shape[0], 2*2*nelem,2*2*nelem))[nnlist]
|
||
leftmatrixlist_TE = leftmatrixlist_s[np.ix_(np.arange(leftmatrixlist_s.shape[0]),TEč,TEč)]
|
||
leftmatrixlist_TM = leftmatrixlist_s[np.ix_(np.arange(leftmatrixlist_s.shape[0]),TMč,TMč)]
|
||
#svarr = np.linalg.svd(leftmatrixlist_TE, compute_uv=False)
|
||
#argsortlist = np.argsort(svarr, axis=-1)[...,:svn]
|
||
#minsvTElist[nnlist] = svarr[...,argsortlist]
|
||
#minsvTElist[nnlist] = np.amin(np.linalg.svd(leftmatrixlist_TE, compute_uv=False), axis=-1)
|
||
minsvTElist[nnlist] = np.linalg.svd(leftmatrixlist_TE, compute_uv=False)[...,-svn:]
|
||
#svarr = np.linalg.svd(leftmatrixlist_TM, compute_uv=False)
|
||
#argsortlist = np.argsort(svarr, axis=-1)[...,:svn]
|
||
#minsvTMlist[nnlist] = svarr[...,argsortlist]
|
||
#minsvTMlist[nnlist] = np.amin(np.linalg.svd(leftmatrixlist_TM, compute_uv=False), axis=-1)
|
||
minsvTMlist[nnlist] = np.linalg.svd(leftmatrixlist_TM, compute_uv=False)[...,-svn:]
|
||
minsvTMlistlist.append(minsvTMlist)
|
||
minsvTElistlist.append(minsvTElist)
|
||
|
||
|
||
|
||
|
||
|
||
# In[ ]:
|
||
|
||
minsvTElistarr = np.array(minsvTElistlist)
|
||
minsvTMlistarr = np.array(minsvTMlistlist)
|
||
omegalist = np.array(omegalist)
|
||
# order to make the scatter plots "nice"
|
||
omegaorder = np.argsort(omegalist)
|
||
omegalist = omegalist[omegaorder]
|
||
minsvTElistarr = minsvTElistarr[omegaorder]
|
||
minsvTMlistarr = minsvTMlistarr[omegaorder]
|
||
|
||
omlist = np.broadcast_to(omegalist[:,nx], minsvTElistarr[...,0].shape)
|
||
kxmlarr = np.broadcast_to(kxmaplist[nx,:], minsvTElistarr[...,0].shape)
|
||
klist = np.concatenate((k0Mlist,kMK1list,kK10list,k0K2list,kK2Mlist), axis=0)
|
||
|
||
|
||
# In[ ]:
|
||
for minN in reversed(range(svn)):
|
||
f, ax = plt.subplots(1, figsize=(20,15))
|
||
sc = ax.scatter(kxmlarr, omlist/eV*hbar, c = np.sqrt(minsvTMlistarr[...,minN]), s =40, lw=0)
|
||
ax.plot(kxmaplist, np.linalg.norm(klist,axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B2-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B2+B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B2-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B2+B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B2-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B1-B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B1-2*B2, axis=-1)*cdn/eV*hbar, '-',
|
||
# kxmaplist, np.linalg.norm(klist+2*B2-B1, axis=-1)*cdn, '-',
|
||
# kxmaplist, np.linalg.norm(klist+2*B1-B2, axis=-1)*cdn, '-',
|
||
)
|
||
ax.set_xlim([np.min(kxmlarr),np.max(kxmlarr)])
|
||
#ax.set_ylim([2.15,2.30])
|
||
ax.set_ylim([np.min(omlist/eV*hbar),np.max(omlist/eV*hbar)])
|
||
ax.set_xticks([0, kxmaplist[len(k0Mlist)-1], kxmaplist[len(k0Mlist)+len(kMK1list)-1], kxmaplist[len(k0Mlist)+len(kMK1list)+len(kK10list)-1], kxmaplist[len(k0Mlist)+len(kMK1list)+len(kK10list)+len(k0K2list)-1], kxmaplist[len(k0Mlist)+len(kMK1list)+len(kK10list)+len(k0K2list)+len(kK2Mlist)-1]])
|
||
ax.set_xticklabels(['Γ', 'M', 'K', 'Γ', 'K\'','M'])
|
||
f.colorbar(sc)
|
||
|
||
pdf.savefig(f)
|
||
|
||
|
||
# In[ ]:
|
||
|
||
f, ax = plt.subplots(1, figsize=(20,15))
|
||
sc = ax.scatter(kxmlarr, omlist/eV*hbar, c = np.sqrt(minsvTElistarr[...,minN]), s =40, lw=0)
|
||
ax.plot(kxmaplist, np.linalg.norm(klist,axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B2-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B2+B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-B2-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist+B2+B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B2-B1, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B1-B2, axis=-1)*cdn/eV*hbar, '-',
|
||
kxmaplist, np.linalg.norm(klist-2*B1-2*B2, axis=-1)*cdn/eV*hbar, '-',
|
||
# kxmaplist, np.linalg.norm(klist+2*B2-B1, axis=-1)*cdn, '-',
|
||
# kxmaplist, np.linalg.norm(klist+2*B1-B2, axis=-1)*cdn, '-',
|
||
)
|
||
ax.set_xlim([np.min(kxmlarr),np.max(kxmlarr)])
|
||
#ax.set_ylim([2.15,2.30])
|
||
ax.set_ylim([np.min(omlist/eV*hbar),np.max(omlist/eV*hbar)])
|
||
ax.set_xticks([0, kxmaplist[len(k0Mlist)-1], kxmaplist[len(k0Mlist)+len(kMK1list)-1], kxmaplist[len(k0Mlist)+len(kMK1list)+len(kK10list)-1], kxmaplist[len(k0Mlist)+len(kMK1list)+len(kK10list)+len(k0K2list)-1], kxmaplist[len(k0Mlist)+len(kMK1list)+len(kK10list)+len(k0K2list)+len(kK2Mlist)-1]])
|
||
ax.set_xticklabels(['Γ', 'M', 'K', 'Γ', 'K\'','M'])
|
||
f.colorbar(sc)
|
||
|
||
pdf.savefig(f)
|
||
pdf.close()
|
||
|
||
print(time.strftime("%H.%M:%S",time.gmtime(time.time()-begtime)))
|